as everybody knows, lame 3.90.3 was tested thorouhly. but this was years ago.
and those tests proved that it was better than 3.89.

then developement took place and lame changed
now a days very few are doing tests. so standard of thorough test should also be changed

we know that 3.90.3 is transparent to most.
but if these new alphas are also transparent to most, then how can we say that which one is superior? along with problem samples, real music must also be taken into account.

lets have it like this:
we can say standard is transparent, no need to go upto extreme or insane though they are superior. (i've seen posts like this)
then why we cant say 397a is transparent as well as fast, no need to go back to 3.90 though it is tested more.

we can leave V0 or - 320 because we have V2 (2 is lower compared to 0, but we cannot distinguish as it is very high itself) as standard
then why cant we leave 3.90 because we have 397a vbr new?

afterall its lossy compression! every setting is a tradeoff size/qual. as long as we get transparent/optimum/reasonable o/p we usually spare higher settings, thouh they are giving more quality technically, or by test. then why cant we accept a new fact which is also transparent. no matter if it might be a little less in quality (just because its not that tested!) but still it is giving great advantages like speed and reduced size!

it's just a thought. no offence intended. just a thought developed in mind after reading so many articles about all these versions.

lets see howmany of us are approving 397 alpha and wish that it becomes stable as soon as possible.

I'm splitting my usage between 3.97a11 (using --vbr-new) and 3.96.1 (using the default old vbr). I encode at -V2, -V3 and -V4. I'm looking forward to final 3.97 being released, though I'm not in any huge need of hurry. Thanks to all the Lame devs for their work on the project.

3.97 alpha : solves most "background ringing" issue at low/mid bitrate with ABR/CBR.

P.S. There are also two similar polls:

- Upgrade the official HA LAME version?, Now (3.96) or wait until 4.0?
- Which MP3 codec do you prefer?, LAME 3.90.3, 3.96.1, or something else?

I have ripped and compressed all of my CD collection. I will not re-rip and recompress unless I am snowed in for a month. ;o) I will wait for 3.97 to be blessed as good enough for general use. Until then I am using 3.96.1.

Technically 3.96.1 but I've started testing 3.97 alpha 11. I'm thinking once 3.97 gets to beta (which probably won't be very long from now) I'll go ahead and move up

Split decision for me, i use 3.90.3 for APS, but for lower bitrates, 3.97a11 is excelent.

right ^^

i only use lame at APS (when i use it at all)

in those cases i use 3.90.3. if i ever needed a low {C,A}BR i'd use lame from cvs


later

3.97 for all mp3 encoding.

3.97a11 --vbr-new from -V3 through -V5 (generally it's -V4 which is nearly a perfect quality/size setting with 3.97 alphas). Excellent tuning, I'd say.

Mainly: 3.97a11 '-V4 --vbr-new' for portable use.
I'll use 3.97a11 '-V2 --vbr-new' for torrent sharing (of non commercial material like radio programs, promo CDs etc).

I have been a 3.90.3 zealot for a long time and never liked what I saw from the testing threads of 3.96.1. It just wasnt good enough to be a successor to the king. But I won't get into any arguments with anyone about that now

From what I've seen of Guru's and others testing I think this is the version of lame we have been waiting for.

I'm encoding my music with 3.97 --preset standard --vbr-new.

I've switched to 3.97a11 --alt-preset standard --vbr-new, recently.

great reply friends!
keep going!
nice to see popularity of 3.97
i think this is going to be last of lame 3.xx family
and hope it will be accepted as the best globally.

faster, smaller, about 180 kbps.

But I've noticed the speed is significantly slower in alpha 11 than previous alpha 8.
Do u guys experience this too?

I'm going to stick with my Lame 3.96.1 untill they come out with a beta for this release. Not a big fan of re-encoding my CDs and I don't have enough space for keeping the uncompressed wav files handy.
J

... that's what lossless audio compression is for

flac, wavpack, etc.


later

I have grudgingly switched from 3.90.3 APS to 3.97 V2 vbr new. While I would rather use a stable release, the reports regarding sound quality coming in from golden eared listeners has convinced me to switch. Smaller file sizes and faster encoding vs 3.90.3 is a nice bonus as well.

Those formats are nice, but they don't really meet my needs. For one they're a bit too large for my portable device and they are eating up almost as much HD space as the original wav file.

I'd rather just keep it in wav format than another format that is only saving me a small amount of space. At least with the wav file I can encode it any time I desire into a different format.
J

dude... 60% compression isn't exactly "a small amount" ??? (i.e. 40% reduction)

lets say you have 10 gigs of uncompressed wavs. in almost any lossless format that is now about 6 gigs. also, most lossless formats are easy to transcode to anything else.


later

so! once again it seems that 3.97 is the most popular one amongst the community. and most of the others are using the latest stable lame. only 14% using 3.90.3.
thanx to guruboolez for excellent testing done on latest alphas. and to gabriel for implementing lame. and last but not least thanks to Darin Morrison for starting this whole quality concept, after which lame became the best mp3 encoder on the planet.

as is the case with me. 3.97 alpha 11 for all my encodings. i used -V0 first. file sizes are smaller than with lame 3.96 APS. But the new alpha seems to be tuned particularly for -V2 --vbr-new, so i switched to that. since i have a lot of new classical music to discover yet, i think the smaller file sizes now come in handy.
So much so that i have reencoded all my old musepack APS to this format. And it works especially well for classical music as guruboolez himself has noted.

gabriel or guru! can you confirm this thing? is it true that v2 is more tuned than v0?

and another question. in older times (3.90), APS used to enable some special codes that couldnt be enabled by switches.
but new lame itself is special. all alternate preset code is merged in mainstream and even without using preset we can use that.
so from now can we consider that only V2 is not the "bible standard"? and consider that V0 will be more effective than V2 in terms of quality?
(considering we are using vbr-new only)
or this V2 still stands most tuned and king of all?

I'm not working on LAME, and therefore I can't answer.

It is true that there had been less listening test results regarding V0 than regarding V2, but V0 should still sound superior to V2.

I'm using 3.96.1 untill 3.97 gets out of the alpha stage. I have most of my collection stored in lossless on my HD so reencoding won't be too difficult.

Funny that only about 15% of the members that voted actually use the recommended encoder version.

Quite a disobedient crowd, huh?

I was thinking the same thing. It makes sense for the recommended version to be conservative, I guess, but HA's recommdation has been pretty darn conservative. However, this is nowhere near what ÜberNet does - they absolutely require all mp3's being traded on their network be 3.90.3, 3.90.2 or 3.92 (the latter two are "grudgingly accepted"). 3.96.1 isn't tested enough to be trusted.
If you want to read over their exacting standards for ripping, check out
http://www.ubernet.org/?p=UberStandard
It's quite interesting, and there are actually points at which I specifically do things differently than they recommend, like trimming silence (using wavtrim) off the beginning of songs, or doing different replaygain stuff to optimize gains per song (best-of albums, especially, often have some songs that are much louder than others because the mastering wasn't standardized, and I sometimes go through and gain them to approximately equal *listening* volume, which isn't necessarily equal numerical gain).

Regarding -V0 being less tuned than -V2.

Frankly, this has always been the case. It is impossible to tune a preset that aims for "better than transparent". -V0 (or --preset extreme) has actually never been proven to be better at anything, but just throws more bandwidth at the same problems resulting in a THEORETICAL benefit. As far as I know, problem samples are still problematic with -V0.

Any problem samples discovered are solved (or an attempt is made to solve them) at -V2, so yes - that is where all the tuning takes place, and has taken place since the 3.90 days.

EDIT: I should clarify though that improvements to -V2 benefit -V0 as well, so at the very least -V0 will be equivalent in quality to -V2 in any given version. So in this sense, -V0 IS improving too.

They should list Ubernet in dictionary definition for anal...

What's the difference between "-V2 --vbr-new" and "--preset standard --vbr-new"? By themselves they (-V2 and -aps) generate the same files, but with --vbr-new they generate different ones, so which one should people actually be using, since I've seen numerous posts with either one.

Edit: clarify: not asking whether people should be using --vbr-new or not, but whether it should go with '-V2' or '-aps', as they generate different files when used in combination with '--vbr-new'.

They should be the same. The easier way to write --preset standard --vbr-new is --preset fast standard


-V2 is equal to --preset standard
-V2 --vbr-new is equal to --preset fast standard
The fast/new vbr algorithm has been developing over time, and now in the 3.97 alphas, is finally equal to or exceeding the original (slower) vbr algorithm in quality.

This is not entirely correct. Sure big problems in V2 are still audiable in V0, but smaller ones are either reduced or become inaudiable. I've managed to abx some artifacts on V2 here and there - little differences, some were a little annoying. I had lots of 7/8 results on normal music. When I pushed to 16 trials I managed 13 or 14/16 at times. It wasn't easy and I had to pause a lot.. Anyway I had no such luck on V0 ever - not with 3.90.3, 3.96.1 , 3.97 , vbr new or old.

Yes, they should be the same, but aren't. I know that -aps --vbr-new aliases to -apfs, but -V2 --vbr-new and --aps --vbr-new generate different data and histograms.

Edit: -V2 --vbr-new generates the same files as --alt-preset fast standard, which is expected because of the mappings. However, these two files are different from --alt-preset standard --vbr-new, which is kind of odd.

i think gabriel stated why this happens.
--aps --vbr-new immediately runs --aps and doesnt take --vbr-new into account

That isn't exactly correct --aps --vbr-new does take --vbr-new into account, but at the wrong moment:



link to Gabriel's post

indeed, you should use "--preset fast standard", not "--preset standard --vbr-new"

The newer,the better?

Not always. But in this case, yes.

Would these comments about -V0 also apply to -V1 ?

Regards,
Madrigal

Currently I'm using 3.96.1, only because I'm afraid of the 'alpha' tag. I'll probably re-rip my collection of 3.97 when it goes final (hopefully soon ).

i am using lame 3.97a11 for transcoding my mp3s into 128kbps cbr mp3s (it's required for my portable). and i like the the way it sounds.

I'm a tad apprehensive regarding alpha software too, therefore I'm still using 3.96.1 until 3.97 beta comes out.

what are your expereinces with bitrate distributons, averaged bitrates following the -V -vbr-new) system in lame 3.97 alpha12 eg. ?

in the 3.95 topic, for -V5 is stated ca. 130 as "target"/averaged bitrate, which I can confirm for 3.97a12, too.

My findings with 3.97a12 go from 102 - 151 even for 1 album, which averages to 130.

And what would be the best quality setting for medium to low bitrates, ie. 128 - 160 kbit.,
-V5, -V6, preset 133, preset 140, 150 whatever (as abr) or better -V5 -vbr-new , compared to -V5, as the -vbr-new setting has been worked out more , so -vbr-new is faster and better quality ?!

From my experience the bitrate distribution in 3.97.x and --vbr-new is very poor compared with old vbr method. Often you have like 50% at one bitrate, 25 at both bitrates around.

This does not say anything about quality. But makes me sceptic...

i encoded around 50 songs with 3.97a10 -V4 --vbr-new and avg bitrate is around 150k. Can't tell much from 3 samples.

I've encoded about 1100 songs (~6 GBs) in various genres (complex electronics, progressive rock, ambient, industrial, etc.) with that commandline and the average bitrate is 150 kbps (actually, some albums went, ~100kbps and they sound pretty well even with that ridiculous bitrate). I think this value can be safely considered as a target bitrate for --(fast) medium preset.

So it does, for some reason, bits seem to cluster around a particular value. Still the file size is smaller, the sound is better. i don't get it. The phenomenon does occur at V2 to V0, at the standard settings.
Could one of the lame developers comment on this? Is it that the bitrate distribution needs further tuning or what?

Do not look at the bitrate distribution if it is annoying you.
More seriously, frame size distribution is not that relevant. Even using the same algorithm, frame size distribution could be made quite different while preserving the same number of bits per frame by a different use of the reservoir.

If mp3 repacker was used on the two files, would the output then be the same?